A smart equipment maintenance method and terminal based on big data and physical ID

Abstract

The present invention discloses an intelligent equipment maintenance method and terminal based on big data and physical ID. Before the equipment maintenance is carried out, the device to be repaired is obtained by scanning the physical ID of the device, and the corresponding historical defect information is searched according to the information of the device to be repaired. The defect score of the device to be repaired is then performed based on the defect measurement model, so that the equipment problem can be quickly located. During the equipment maintenance process, the historical defect information of the device to be repaired is semantically analyzed, and defect treatment opinions are associated with the device to be repaired. After the equipment maintenance, a maintenance model is established based on the defect score and defect treatment opinions of the device to be repaired, and a maintenance plan for the device to be repaired is generated according to the maintenance model. Therefore, the located device can be automatically detected for defects and matched with corresponding solutions, thereby automatically and intelligently performing equipment maintenance.

Description

Technical Field

[0001] The present invention relates to the technical field of maintenance of electric ultra-high voltage equipment, and in particular to an intelligent equipment maintenance method and terminal based on big data and physical ID. Background Art

[0002] As a vital component of the power grid, substation stability is crucial to the safe operation of the entire system. Failures can easily lead to grid accidents. Inadequate maintenance of substation equipment and unchecked hidden dangers often lead to substation failures. Therefore, timely maintenance of substation equipment is crucial.

[0003] In the current State Grid Corporation's PMS (equipment management system), during the substation maintenance process, operation and maintenance personnel are required to enter the substation, inspect each device one by one, and then judge the equipment status based on their experience. At the same time, a maintenance report needs to be prepared during the maintenance process and entered into the PMS terminal after the maintenance is completed.

[0004] The maintenance process involves a wide variety of equipment, with hundreds of different types and manufacturers, each with its own unique parameters. Identifying certain defects and potential hazards often relies on the technical expertise of the maintenance personnel. In actual operation and maintenance, the varying skills of maintenance personnel can lead to different conclusions when inspecting the same equipment.

[0005] Therefore, there is an urgent need for an intelligent analysis and judgment terminal that relies on big data. On the one hand, it can automatically perform preliminary defect identification based on artificial intelligence technology. On the other hand, it can automatically provide a defect gallery of similar equipment based on the equipment type for reference by maintenance personnel, thereby achieving standard unification in the equipment maintenance process. Summary of the Invention

[0006] The technical problem to be solved by the present invention is to provide an intelligent equipment maintenance method and terminal based on big data and physical ID, which can quickly locate equipment problems and match corresponding solutions.

[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0008] A method for intelligent equipment maintenance based on big data and physical ID, comprising the following steps:

[0009] Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired;

[0010] Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics;

[0011] Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired;

[0012] A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

[0013] In order to solve the above technical problems, another technical solution adopted by the present invention is:

[0014] A smart equipment maintenance terminal based on big data and physical ID includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0015] Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired;

[0016] Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics;

[0017] Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired;

[0018] A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

[0019] The beneficial effects of the present invention are as follows: before equipment maintenance is carried out, the device to be repaired is obtained by scanning the physical ID of the device to be repaired, the corresponding historical defect information is searched based on the information of the device to be repaired, and the defect score of the device to be repaired is performed based on the defect measurement model, so that equipment problems can be quickly located. During the equipment maintenance process, the historical defect information of the device to be repaired is semantically analyzed, and defect treatment opinions are associated with the device to be repaired. After the equipment maintenance, a maintenance model is established based on the defect score and defect treatment opinions of the device to be repaired, and a maintenance plan for the device to be repaired is generated based on the maintenance model. Therefore, it is possible to automatically detect defects in the located equipment and match corresponding solutions, thereby automatically and intelligently performing equipment maintenance. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 This is a flow chart of an intelligent equipment maintenance method based on big data and physical ID according to an embodiment of the present invention;

[0021] Figure 2 Schematic diagram of an intelligent equipment maintenance terminal based on big data and physical ID according to an embodiment of the present invention;

[0022] Figure 3 This is a module diagram of an intelligent equipment maintenance method based on big data and physical ID according to an embodiment of the present invention;

[0023] Description of labels:

[0024] 1. An intelligent equipment maintenance terminal based on big data and physical ID; 2. Memory; 3. Processor. DETAILED DESCRIPTION

[0025] To illustrate the technical content, achieved objectives and effects of the present invention in detail, the following description is given in conjunction with the embodiments and accompanying drawings.

[0026] Please refer to Figure 1 The embodiment of the present invention provides a method for intelligent equipment maintenance based on big data and physical ID, including the following steps:

[0027] Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired;

[0028] Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics;

[0029] Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired;

[0030] A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

[0031] As can be seen from the above description, the beneficial effects of the present invention are as follows: before equipment maintenance is carried out, the equipment to be repaired is obtained by scanning the physical ID of the equipment to be repaired, the corresponding historical defect information is searched based on the information of the equipment to be repaired, and the defect score of the equipment to be repaired is performed based on the defect measurement model, so that equipment problems can be quickly located. During the equipment maintenance process, the historical defect information of the equipment to be repaired is semantically analyzed, and defect handling opinions are associated with the equipment to be repaired. After the equipment maintenance, a maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated based on the maintenance model. Therefore, it is possible to automatically detect defects on the located equipment and match corresponding solutions, thereby automatically and intelligently performing equipment maintenance.

[0032] Furthermore, using a defect measurement model to analyze the historical defect information of the equipment to be repaired and scoring the defect of the equipment to be repaired based on data analysis and data statistics includes:

[0033] Preprocessing the historical defect information of the equipment to be repaired, and performing data analysis and statistics on the preprocessed data;

[0034] Variables required for the defect measurement model are screened, and after the defect measurement model is established based on Woe binning, a defect score card is generated according to the Woe value, and the maintenance equipment is scored based on the defect score card.

[0035] From the above description, it can be seen that by obtaining the score of the equipment to be repaired through the defect measurement model and the defect score card, the status of the equipment and its defects can be understood.

[0036] Furthermore, the defect scoring of the equipment to be repaired based on data analysis and data statistics also includes:

[0037] Score the equipment to be repaired and its defects respectively, preset the color value array, and calculate the color values ​​of the equipment to be repaired and its defects:

[0038] Defect color value = color value array * defect score coefficient;

[0039] Device color value = color value array * device rating coefficient;

[0040] The equipment to be repaired and its defect information are visually displayed based on the color values ​​of the equipment to be repaired and its defects.

[0041] As can be seen from the above description, the color-based visual display of defect scores and device scores can more intuitively understand the status of the device and its defects.

[0042] Furthermore, semantic analysis is performed on the historical defect information of the equipment to be repaired, and corresponding defect handling opinions are associated with the equipment to be repaired, including:

[0043] Classifying the historical defect information of the equipment to be repaired through semantic analysis, establishing a category tree based on the category tags obtained by the classification, and associating the historical information of the equipment to be repaired based on the category tree;

[0044] A convolutional neural network algorithm is used to match the historical information and historical defect information of the equipment to be repaired to obtain corresponding defect handling opinions.

[0045] From the above description, it can be seen that the use of intelligent semantic analysis algorithms can achieve parallelization of data and topic models, and can obtain corresponding processing opinions based on defect information analysis.

[0046] Furthermore, a maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The maintenance plan is used to repair the scanned equipment to be repaired, including:

[0047] Analyze the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history, and establish and store the maintenance model;

[0048] Receive the scanned equipment to be repaired, perform maintenance calculation according to the maintenance model, regularly query the maintenance plan of the equipment to be repaired calculated by the maintenance model based on the calculation results, and actively report the maintenance needs of the equipment to be repaired.

[0049] From the above description, it can be seen that by analyzing the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history and building a model, it is possible to discover maintenance plans that meet the equipment defects and automatically report them.

[0050] Please refer to Figure 2 Another embodiment of the present invention provides an intelligent equipment maintenance terminal based on big data and physical ID, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the following steps are implemented:

[0051] Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired;

[0052] Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics;

[0053] Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired;

[0054] A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

[0055] As can be seen from the above description, before equipment maintenance begins, the device to be repaired is obtained by scanning the device's physical ID. Based on the device's information, the corresponding historical defect information is searched, and a defect score is assigned to the device to be repaired based on a defect measurement model, enabling rapid identification of equipment issues. During the equipment maintenance process, semantic analysis is performed on the device's historical defect information, and defect resolution suggestions are associated with the device to be repaired. After the equipment maintenance is complete, a maintenance model is established based on the device's defect score and defect resolution suggestions. A maintenance plan for the device to be repaired is generated based on the maintenance model. This allows for automatic defect detection of the located device and matching of appropriate solutions, enabling automated and intelligent equipment maintenance.

[0056] Furthermore, using a defect measurement model to analyze the historical defect information of the equipment to be repaired and scoring the defect of the equipment to be repaired based on data analysis and data statistics includes:

[0057] Preprocessing the historical defect information of the equipment to be repaired, and performing data analysis and statistics on the preprocessed data;

[0058] Variables required for the defect measurement model are screened, and after the defect measurement model is established based on Woe binning, a defect score card is generated according to the Woe value, and the maintenance equipment is scored based on the defect score card.

[0059] From the above description, it can be seen that by obtaining the score of the equipment to be repaired through the defect measurement model and the defect score card, the status of the equipment and its defects can be understood.

[0060] Furthermore, the defect scoring of the equipment to be repaired based on data analysis and data statistics also includes:

[0061] Score the equipment to be repaired and its defects respectively, preset the color value array, and calculate the color values ​​of the equipment to be repaired and its defects:

[0062] Defect color value = color value array * defect score coefficient;

[0063] Device color value = color value array * device rating coefficient;

[0064] The equipment to be repaired and its defect information are visually displayed based on the color values ​​of the equipment to be repaired and its defects.

[0065] As can be seen from the above description, the color-based visual display of defect scores and device scores can more intuitively understand the status of the device and its defects.

[0066] Furthermore, semantic analysis is performed on the historical defect information of the equipment to be repaired, and corresponding defect handling opinions are associated with the equipment to be repaired, including:

[0067] Classifying the historical defect information of the equipment to be repaired through semantic analysis, establishing a category tree based on the category tags obtained by the classification, and associating the historical information of the equipment to be repaired based on the category tree;

[0068] A convolutional neural network algorithm is used to match the historical information and historical defect information of the equipment to be repaired to obtain corresponding defect handling opinions.

[0069] From the above description, it can be seen that the use of intelligent semantic analysis algorithms can achieve parallelization of data and topic models, and can obtain corresponding processing opinions based on defect information analysis.

[0070] Furthermore, a maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The maintenance plan is used to repair the scanned equipment to be repaired, including:

[0071] Analyze the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history, and establish and store the maintenance model;

[0072] Receive the scanned equipment to be repaired, perform maintenance calculation according to the maintenance model, regularly query the maintenance plan of the equipment to be repaired calculated by the maintenance model based on the calculation results, and actively report the maintenance needs of the equipment to be repaired.

[0073] From the above description, it can be seen that by analyzing the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history and building a model, it is possible to discover maintenance plans that meet the equipment defects and automatically report them.

[0074] The above-mentioned intelligent equipment maintenance method and terminal based on big data and physical ID of the present invention are suitable for intelligent equipment maintenance, can quickly locate equipment problems and match corresponding solutions, and are described below through specific implementation methods:

[0075] Example 1

[0076] Please refer to Figure 1 and Figure 3 , a method for intelligent equipment maintenance based on big data and physical ID, comprising the steps of:

[0077] S1. Obtain the device to be repaired by scanning the physical ID of the device, and search for corresponding historical defect information based on the information of the device to be repaired.

[0078] Specifically, maintenance personnel enter the substation with a work permit and begin deploying safety measures. They use the app to scan and decode the QR codes on various substation devices, query the system for device and parameter information, and display the device information associated with the substation, thereby obtaining the device to be inspected, as scanned by the app.

[0079] The information of the equipment to be repaired can be intelligently associated with the previous defects of the equipment, and the information of the equipment defects can be viewed.

[0080] S2. Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics.

[0081] Specifically, based on defect history, a defect measurement model and defect scorecard are used. Each defect is scored through exploratory data analysis and descriptive statistics, and then a comprehensive score is generated for the equipment, providing a more intuitive understanding of the status of the equipment and its defects. This algorithm model has the advantages of requiring less data and less demanding computer hardware. Compared to other scoring schemes, it achieves a 93% improvement in overall efficiency and a 52% reduction in hardware performance requirements.

[0082] S3. Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired.

[0083] Specifically, for the information of the equipment, the parallel LDA (Latent Dirichlet Allocation) intelligent semantic analysis algorithm is used to realize the parallelization of data and topic models. Through data cleaning, word segmentation, part-of-speech tagging, named entity recognition, word vector, syntactic semantic dependency analysis, similarity algorithm, text classification task, text generation task and other steps, historical defect handling reports, equipment drawings, manuals, historical experience database, auxiliary repair database and other materials and expert opinions are associated with the defect information. At the same time, the expert system can provide corresponding handling opinions for the defects based on the characteristics of the defect information. Quick query and viewing are supported on the APP side. The parallel LDA algorithm mainly used in this embodiment has significantly improved the efficiency by more than 20% compared with the method of qualitatively characterizing defects based on multi-dimensional information.

[0084] In some embodiments, after equipment maintenance is completed, the maintenance test report is directly placed under a designated camera. Customized OCR (Optical Character Recognition) technology automatically reads the recorded information, automatically fills in the maintenance test report template, and stores it in the PMS system for easy access and reading. Information generated after repairing substation equipment defects can be easily recognized through OCR technology to form standardized, digitized reports and content, which can be easily accessed later during the equipment maintenance process. Directly generating digitized content eliminates the process of filling out reports, copying, and archiving, and allows for quick query and access in the future.

[0085] S4. Establish a maintenance model based on the defect score and defect handling opinions of the equipment to be repaired, generate a maintenance plan for the equipment to be repaired according to the maintenance model, and use the maintenance plan to repair the scanned equipment to be repaired.

[0086] Specifically, in daily work after the equipment maintenance is completed, the system applies data modeling based on the limited information provided by the planning system to establish an equipment demand model, including three dimensions: the number of defects, the severity of defects, and the measures required for defects. The Apriori association algorithm automatically runs the data model to discover maintenance plans that meet the equipment defects and automatically reports them. After scanning the "physical ID" of the equipment, the system reflects the degree of maintenance demand of the equipment based on the model calculation results: when the planning system monitors a plan that meets the working conditions, it will report the demand based on the demand model results. The advantage of using this method is that it breaks down the data walls between different systems and can be applied without affecting the security of the existing systems.

[0087] Therefore, this embodiment generates a corresponding detection plan by scanning the equipment to be detected, which greatly improves the efficiency and speed of defect discovery and matching of relevant data during the maintenance process, while greatly reducing the requirements for hardware; and greatly improves work efficiency and reduces time costs. Taking the survey work as an example, for the common routine inspection and survey of line intervals, full preparation should be made for switches, knife switches, secondary drawings and maintenance manuals, CVT (Capacitance type voltage transformer) secondary wiring diagrams, etc., which requires half a day to a day of work. After adopting the maintenance method of this embodiment, the same routine inspection and survey of line intervals can be completed in only 1 to 2 hours, and the working time reduction effect is significant.

[0088] Furthermore, this embodiment improves planning coverage, avoiding repeated outages and redundant personnel dispatch. For example, a routine inspection of an interval requires one vehicle, one driver, two maintenance personnel, and one day's work. This can be avoided by implementing the inspection method of this embodiment. By integrating this into routine inspections, rectification can be completed in only about an hour.

[0089] Furthermore, this embodiment can promote digital maintenance and reduce paper usage. On-site drawings, manuals, and test reports often require hundreds of pages of A4 paper. If a substation performs 100 routine inspections annually, this translates to tens of thousands of pages of paper consumed, in addition to the toner and printer wear and tear costs. By implementing this system, digital maintenance can be implemented, completely eliminating these costs. Annually, this saves 10,000 sheets of A4 paper and four boxes of toner, equivalent to over 5,000 RMB per station.

[0090] Example 2

[0091] This embodiment differs from the first embodiment in that the method for calculating the defect score is further limited. Specifically:

[0092] Preprocessing the historical defect information of the equipment to be repaired, and performing data analysis and statistics on the preprocessed data;

[0093] Screen the variables required for the defect measurement model, establish the defect measurement model based on Woe binning, generate a defect score card according to the Woe value, and score the maintenance equipment based on the defect score card;

[0094] Score the equipment to be repaired and its defects respectively, preset the color value array, and calculate the color values ​​of the equipment to be repaired and its defects:

[0095] Defect color value = color value array * defect score coefficient;

[0096] Device color value = color value array * device rating coefficient;

[0097] The equipment to be repaired and its defect information are visually displayed based on the color values ​​of the equipment to be repaired and its defects.

[0098] In this embodiment, the defect risk measurement model includes defect history rating, overall equipment rating, and overall line rating. The defect rating is composed of a series of rating models, including A card (equipment score card), B card (historical defect model), C card (line model), and F card (tracking model).

[0099] The main development process of a typical defect scoring card model is as follows:

[0100] 1. Obtain data, including historical data on equipment defects. This data includes all dimensions of defects, including frequency, severity, impact, speed of defect elimination, and difficulty of defect elimination.

[0101] 2. Data preprocessing: The main tasks include data cleaning, missing value processing, outlier processing, data type conversion, etc., which requires converting the original data layer by layer into modelable data.

[0102] 3. EDA exploratory data analysis and descriptive statistics, including statistics on the size of the overall data, defect level ratio, data types, variable missing rate, variable frequency analysis histogram visualization, box plot visualization, variable correlation visualization, etc.

[0103] 4. Variable selection: Use statistical and machine learning methods to identify the variables that most significantly influence default status. Numerous variable selection methods exist, including IV, feature importance, and variance. Variables with high missing rates, no business explanatory value, and no value should also be removed.

[0104] 5. Model development. The main challenges in scorecard modeling are Woe binning, score stretching, and variable coefficient calculation. Woe binning is particularly challenging within scorecards, requiring extensive statistical knowledge and business experience. Currently, there are over 50 binning algorithms, with no unified standard. Typically, binning is performed automatically by a machine, followed by manual adjustments. Finally, the model's performance is repeatedly tested to select the optimal binning algorithm.

[0105] 6. Model validation: Verify the model's ability to distinguish, predict, stabilize, and rank, and generate a model evaluation report to determine whether the model is viable. Model validation isn't a one-time process; it's performed regularly after modeling, before launch, and after launch. Model development and maintenance is a cyclical process, not a one-time completion.

[0106] 7. Defect Scorecard: This scorecard is generated based on the logistic regression variable coefficients and Woe values. This scorecard is easy to interpret, very stable, and well-received by the industry. It converts the logistic regression model probability score into a standard score of 1-9.

[0107] 8. Establish a scoring card model system and establish a computer-automated credit scoring system based on the credit scoring card method.

[0108] After scoring defects and equipment, the system visualizes the data based on EDA (Exploratory Data Analysis) and descriptive statistics, marking them with different colors. By applying the device and defect status colors in the list, you can quickly find and view the corresponding equipment and defects.

[0109] In this example, EDA and descriptive statistics include statistics on overall data size, defect level percentage, data type, variable missing rate, variable frequency analysis, histogram visualization, box plot visualization, and variable correlation visualization. Common exploratory data analysis methods include histograms, scatter plots, boxer plots, heat maps, and paired plots.

[0110] Therefore, this embodiment conducts in-depth statistics and analysis of substation equipment and its defects, and categorizes defects based on their status and severity. The system automatically scores each defect based on defect history using a defect measurement model and defect scorecard, using EDA exploratory data analysis and descriptive statistics. This system then provides a comprehensive score for each device, providing a more intuitive understanding of the equipment and its defect status. This algorithm model has the advantages of requiring minimal data and low computer hardware requirements. Compared to other scoring schemes, it achieves a 93% improvement in overall efficiency and a 52% reduction in hardware performance requirements.

[0111] Example 3

[0112] This embodiment differs from the first and second embodiments in that the method for generating defect handling suggestions is further limited. Specifically:

[0113] Classifying the historical defect information of the equipment to be repaired through semantic analysis, establishing a category tree based on the category tags obtained by the classification, and associating the historical information of the equipment to be repaired based on the category tree;

[0114] A convolutional neural network algorithm is used to match the historical information and historical defect information of the equipment to be repaired to obtain corresponding defect handling opinions.

[0115] In this example, equipment defect information and parameters are first analyzed using big data to identify semantic combinations and frequency relationships, and then perform text classification. For example, for a defective device called "220kV imitation long-line III circuit 249 unit combination electrical appliance," the most frequently occurring keywords are "220kV," "imitation long-line III circuit," "249," and "combination electrical appliance." These keywords are stored using an intelligent semantic algorithm. Then, information in the historical database, including drawings and manuals, is semantically segmented and stored.

[0116] Text classification is the most common text semantic analysis task. While simple, almost everyone who has worked with NLP has done it. However, it's also complex. Achieving an accuracy rate of over 90% for a text classification task with hundreds of labels per category remains challenging. In this example, text classification refers to general text classification, including novelty search classification, advertisement classification, page classification, and user classification.

[0117] Almost all machine learning methods can be used for text classification, but this embodiment adopts the best implementation of the current algorithm. The steps are as follows:

[0118] 1. Establish a classification system. Due to limited knowledge, a single category system cannot cover all situations. Furthermore, there may be imbalances between categories. Consequently, the relationship between one category and another is uncertain, which can lead to difficulties in use and optimization. Therefore, the category system is manually identified, explored, and established, and then refined based on the existing category system.

[0119] 2. Build a category tree based on the category labels in a certain hierarchical relationship. Using a hierarchical classifier for classification, first train a classifier for the first-layer nodes, then train n classifiers for the second layer (n is the number of nodes in the first layer), and so on. Using a hierarchical category tree makes individual models simpler and more accurate while also avoiding cross-influence between category labels.

[0120] Secondly, by analyzing the information and key descriptions of equipment defects through big data, semantic combinations and frequency relationships are discovered and then segmented and stored using intelligent semantic algorithms. The information in the historical database, including drawings and manuals, is then semantically segmented and stored. By semantically segmenting and storing the titles of the historical experience database and supplementary materials, parallelized LDA inference is performed using the EM (Expectation-Maximization Algorithm) algorithm to derive the local optimal solution. The associated relationships are stored through a program and displayed on the app through program calls, supporting quick query and download for viewing.

[0121] 1. Randomly initialize the topic of each word and count two frequency count matrices: the document topic count matrix N(t,d), which is used to describe the topic frequency distribution in each document; and the word topic count matrix N(w,t), which is used to represent the frequency distribution of words under each topic.

[0122] 2. Traverse the training corpus, resample the topic corresponding to each word according to the probability formula, and update the counts of N(t,d) and N(w,t).

[0123] 3. Repeat step 2 until the model converges.

[0124] The parallelization of topic models, not just topic models, can be explained from two perspectives: data parallelism and model parallelism.

[0125] Data and model parallelism can be described as a chessboard, with rows divided by data and columns by model. LDA parallelization utilizes this partitioning to split the originally large matrix, which would be impossible to store on a single machine, across different machines, allowing each machine to store parameters in memory. Each word vector is then calculated relatively independently, with model data synchronized periodically according to certain strategies.

[0126] In text analysis models, vectors are used to describe a word's "word embedding." The use of word embeddings simplifies the model and truly leverages context for prediction. The basic principles behind word embeddings are: Word embeddings can be used to explore relationships between words, apply word embeddings as features to other machine learning tasks, and even be applied to machine translation. This characteristic can be leveraged to extract hierarchical relationships between words, among other advantages.

[0127] The word vectors corresponding to all the constituent words are summed up to form the phrase vector and sentence vector. These vectors are then analyzed. During training, IDs are added, meaning each sentence in the training corpus has a unique ID. During the prediction phase, a new paragraph ID is assigned to the sentence to be predicted. The parameters of the word vectors and output layer remain unchanged from the training phase, and gradient descent is used again to train the sentence to be predicted. After convergence, the vector for the sentence to be predicted is obtained.

[0128] Then, after semantically segmenting the device type information within the defect database, the corresponding device category is located in the expert database. The semantically segmented text within the defect content within the defect database is combined with the component and problem description columns within the expert database. A CNN-based convolutional neural network algorithm is applied to the text to match all expert opinions for this type of device and defect, allowing for quick review and repair of relevant opinions.

[0129] Convolutional Neural Networks (CNNs) can be used for tasks such as text classification, sentiment analysis, and ontology classification. Traditional text classification tasks typically rely on word meaning or feature extraction based on the word itself. These methods generally require domain knowledge and artificial features. CNNs use similar methods, but are generally based on raw text. CNN model inputs can be word sets, word vectors, or even simple characters. Compared to traditional methods, CNNs do not require as many artificial features.

[0130] A CNN is used to perform text classification using a word set as input. The CNN consists of four layers. The first layer is the word embedding layer, which maps each word in the document to a word embedding space. Assuming a word embedding is k-dimensional, then after mapping n words, it is equivalent to generating an n*k-dimensional image. The second layer is the convolutional layer, where multiple filters are applied to the word embedding layer, and different filters generate different feature maps. The third layer is the pooling layer, which takes the maximum value of each feature map. This operation can handle documents of varying lengths, as the output of the third layer depends only on the number of filters. The fourth layer is a fully connected top-level soft layer, which outputs the probability of each category. Furthermore, the input layer can have two channels: one channel uses word embeddings pre-trained using the parallel LDA algorithm, and the other channel's word embeddings are adjusted during training using a reverse algorithm. As a result, the model achieves ideal results on four of seven commonly used classification evaluation tasks, and performs near state-of-the-art on three others.

[0131] As can be seen from the above description, in the process of acquiring and associating relevant data on equipment and its defects, the system uses big data analysis, text classification methods based on text semantic analysis, parallelized LDA algorithms, and convolutional network-based CNN algorithms, among other intelligent semantic algorithms, to perform semantic statistics, splitting, matching, and storage. Furthermore, it uses database and system methods for storage and display. This achieves logical association between information in different databases.

[0132] Example 4

[0133] The difference between this embodiment and the first to third embodiments is that the method for generating the maintenance plan is further limited. Specifically:

[0134] Analyze the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history, and establish and store the maintenance model;

[0135] Receive the scanned equipment to be repaired, perform maintenance calculation according to the maintenance model, regularly query the maintenance plan of the equipment to be repaired calculated by the maintenance model based on the calculation results, and actively report the maintenance needs of the equipment to be repaired.

[0136] In this embodiment, specifically:

[0137] 1. The system applies data modeling by establishing an equipment demand association model based on the three dimensions of the number of equipment defects, defect severity, and defect-required measures by using the association rule Apriori algorithm.

[0138] 2. After scanning the physical ID of the equipment, the system calculates the results according to the model. When the planning system detects a plan that meets the working conditions, it will proactively report the degree of maintenance needs of the equipment based on the calculation results of the demand model.

[0139] Through data analysis, an equipment demand association model is established based on three dimensions: number of defects, defect severity, and required measures, using the Apriori algorithm. The system automatically calls matching applications based on this data model. After scanning the physical device ID, the system automatically matches the model's calculation results with the retrieved device information. If the planning system detects a plan that meets the working conditions, the demand model results reflect the equipment's maintenance need and proactively match it to the plan. When reviewing defects, the planned demand matching this model is automatically displayed and reported.

[0140] As can be seen from the above description, this invention analyzes historical substation equipment defect information (including the number of major defects, severity, and required remedial measures) and employs the Apriori algorithm, an association rule algorithm, to develop a unique equipment demand model. This model enables the system to automatically match equipment information with the current defect maintenance plan and automatically associate it with the equipment, enabling proactive reporting.

[0141] Association rules are an unsupervised machine learning method used for knowledge discovery, not prediction. Association rule learners do not require pre-labeled training data, as unsupervised learning does not involve a training step. However, this model suffers from the difficulty of model evaluation, and the rationality of the results can generally be determined through business experience.

[0142] The Apriori principle states that if an itemset is frequent, then all its subsets are also frequent. The Apriori principle can help reduce computational complexity. A more common proposition is its converse: if an itemset is infrequent, then all its supersets are also infrequent. This is called the anti-monotonicity or downward closure property of the itemset.

[0143] The specific steps are as follows:

[0144] 1. Generate candidate 1-itemset C1, calculate support, and generate frequent 1-itemset L1 based on minimum support.

[0145] Among them, the collection of items is referred to as an itemset, the number of its elements is the length of the itemset, and an itemset with a length of k is called a k-itemset.

[0146] Itemset support is used to describe the importance of X. For itemset X, count is the number of transactions containing X in transaction set D. The support of itemset X is the probability of itemset X appearing:

[0147]

[0148] 2. Generate candidate 2-itemset C2, calculate support, and generate frequent 2-itemset L2 based on the minimum support.

[0149] 3. Generate candidate 3-itemset C3, calculate support, and generate frequent 3-itemset L3 based on minimum support.

[0150] 4. Generate association rules. The simplest method is to list all non-empty true subsets of each frequent item set, randomly select two of them as the LHS (Left Hand Side) and RHS (Right Hand Side), form association rules, calculate the confidence of each association rule, and delete weak rules.

[0151] Among them, the model is automatically triggered after the system scans the "physical ID" of the equipment. The system regularly stores the results of the model calculation and queries the associated plan calculated by the model. The system monitors the plan that meets the working conditions and actively reports the degree of maintenance demand of the equipment based on the calculation results of the demand model.

[0152] As can be seen from the above description, during daily equipment operation, by analyzing and modeling the number of defects, their severity, and the necessary remedial measures, we can achieve the goal of automatically conducting regular equipment inspections. If the model runs and finds a defect that meets the maintenance plan, it can be automatically reported, significantly reducing the manpower required for observation and analysis during daily equipment operation.

[0153] Example 5

[0154] Please refer to Figure 2 A smart equipment maintenance terminal 1 based on big data and physical ID includes a memory 2, a processor 3, and a computer program stored in the memory 2 and executable on the processor 3. When the processor 3 executes the computer program, each step of a smart equipment maintenance method based on big data and physical ID in any one of embodiments one to four is implemented.

[0155] In summary, the present invention provides an intelligent equipment maintenance method and terminal based on big data and physical ID. Before the equipment maintenance is carried out, by scanning the physical ID code of the equipment, the system can automatically decode and query to obtain equipment information and equipment defect information. Using a defect measurement model and a defect scoring card, each defect is scored through EDA exploratory data analysis and descriptive statistics, so as to more intuitively understand the status of the equipment and its defects. During the equipment maintenance process, the system uses the parallel LDA intelligent semantic analysis algorithm to achieve the parallelization of data and topic models. Through data cleaning, word segmentation, part-of-speech tagging, named entity recognition, word vectors, syntactic semantic dependency analysis, similarity algorithm, text classification task, and text generation task steps, it automatically queries the drawings, manuals, historical experience databases, auxiliary repair databases, and expert opinions of related equipment and its defects, which can be quickly referenced for maintenance. After the equipment maintenance is completed, the system can quickly generate and store a maintenance test report through OCR technology. During daily work after equipment maintenance is completed, the system automatically runs the data model using the limited information provided by the planning system and the Apriori association algorithm to identify and automatically report maintenance plans that meet the equipment defects. By operating the app during the equipment maintenance process, the efficiency of equipment maintenance work can be greatly improved, while also increasing plan coverage, avoiding repeated power outages and redundant personnel dispatch, and promoting the digitalization of equipment maintenance processes and daily management.

[0156] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's description and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for intelligent equipment maintenance based on big data and physical ID, characterized by: Including steps: Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired; Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics: preprocessing the historical defect information of the equipment to be repaired, and performing data analysis and statistics on the preprocessed data; Screen the variables required for the defect measurement model, establish the defect measurement model based on Woe binning, generate a defect score card according to the Woe value, and score the maintenance equipment based on the defect score card; The defect scoring of the equipment to be repaired based on data analysis and data statistics further includes: scoring the equipment to be repaired and its defects respectively, and presetting a color value array to calculate the color value of the equipment to be repaired and its defects: defect color value = color value array Defect score coefficient; device color value = color value array Equipment scoring coefficient; visual display of the equipment to be repaired and its defect information based on the color value of the equipment to be repaired and its defects; Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired: perform text classification on the historical defect information of the equipment to be repaired through semantic analysis, establish a category tree based on the category labels obtained by classification, and associate the historical information of the equipment to be repaired based on the category tree; use a convolutional neural network algorithm to match the historical information of the equipment to be repaired and the historical defect information to obtain corresponding defect handling opinions; A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

2. The intelligent equipment maintenance method based on big data and physical ID according to claim 1 is characterized in that: Establishing a maintenance model based on the defect score and defect handling opinions of the equipment to be repaired, generating a maintenance plan for the equipment to be repaired according to the maintenance model, and repairing the scanned equipment to be repaired using the maintenance plan includes: Analyze the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history, and establish and store the maintenance model; Receive the scanned equipment to be repaired, perform maintenance calculation according to the maintenance model, regularly query the maintenance plan of the equipment to be repaired calculated by the maintenance model based on the calculation results, and actively report the maintenance needs of the equipment to be repaired.

3. An intelligent equipment maintenance terminal based on big data and physical ID, comprising a memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that: When the processor executes the computer program, the following steps are implemented: Obtain the device to be repaired by scanning the physical ID of the device, and search for the corresponding historical defect information based on the information of the device to be repaired; Using a defect measurement model to analyze the historical defect information of the equipment to be repaired, and performing defect scoring on the equipment to be repaired based on data analysis and data statistics: preprocessing the historical defect information of the equipment to be repaired, and performing data analysis and statistics on the preprocessed data; Screen the variables required for the defect measurement model, establish the defect measurement model based on Woe binning, generate a defect score card according to the Woe value, and score the maintenance equipment based on the defect score card; The defect scoring of the equipment to be repaired based on data analysis and data statistics further includes: scoring the equipment to be repaired and its defects respectively, and presetting a color value array to calculate the color value of the equipment to be repaired and its defects: defect color value = color value array Defect score coefficient; device color value = color value array Equipment scoring coefficient; visual display of the equipment to be repaired and its defect information based on the color value of the equipment to be repaired and its defects; Perform semantic analysis on the historical defect information of the equipment to be repaired, and associate corresponding defect handling opinions with the equipment to be repaired: perform text classification on the historical defect information of the equipment to be repaired through semantic analysis, establish a category tree based on the category labels obtained by classification, and associate the historical information of the equipment to be repaired based on the category tree; use a convolutional neural network algorithm to match the historical information of the equipment to be repaired and the historical defect information to obtain corresponding defect handling opinions; A maintenance model is established based on the defect score and defect handling opinions of the equipment to be repaired, and a maintenance plan for the equipment to be repaired is generated according to the maintenance model. The scanned equipment to be repaired is repaired using the maintenance plan.

4. The intelligent equipment maintenance terminal based on big data and physical ID according to claim 3 is characterized in that: Establishing a maintenance model based on the defect score and defect handling opinions of the equipment to be repaired, generating a maintenance plan for the equipment to be repaired according to the maintenance model, and repairing the scanned equipment to be repaired using the maintenance plan includes: Analyze the number of defects, defect scores, and defect handling opinions of the equipment to be repaired in history, and establish and store the maintenance model; Receive the scanned equipment to be repaired, perform maintenance calculation according to the maintenance model, regularly query the maintenance plan of the equipment to be repaired calculated by the maintenance model based on the calculation results, and actively report the maintenance needs of the equipment to be repaired.

Images